基于Faster R-CNN的铝型材表面缺陷识别研究  

Research on Surface Defect Recognition of Aluminum Profile Based on Faster R-CNN

作　　者：邓慧[1] 崔亚飞 DENG Hui;CUI Yafei(School of Intelligent Manufacturing and Construction Engineering,Yongzhou Vocational and Technical College,Yongzhou 425100,Hunan)

机构地区：[1]永州职业技术学院智能制造与建筑工程学院,湖南永州425100

出　　处：《济源职业技术学院学报》2022年第3期59-62,共4页Journal of Jiyuan Vocational and Technical College

基　　金：永州市教育科学规划课题(YJK021A016)。

摘　　要：传统铝型材目标检测算法的准确率较低,严重影响铝型材的美观和质量。本文在Faster R-CNN网络的基础上,用ResNeXt105(残差网络)代替原始VGG16(经典卷积神经网络)提取图像特征,设计了Cascade Faster R-CNN的网络结构,采用FPN(特征金字塔网络)提取多尺度特征图并进行特征图融合。实验结果表明,在2722张图像测试集上,Faster R-CNN模型准确率为62.7%,本网络模型测试准确率达到81.4%,提高了18.7%。故相比于其他网络模型,本文的Faster R-CNN模型具有更强的特征提取能力和泛化能力,为类似小目标检测提高了技术参考。the accuracy of traditional aluminum profile target detection algorithm is low,which seriously affects the beauty and quality of aluminum profile.Based on the Fast R-CNN network,this paper uses the residual network ResNeXt105 to extract the image features instead of the original classical convolutional neural network vgg16,designs the network structure of Cascade Fast R-CNN,and uses the feature pyramid network FPN to extract the multi-scale feature map and fuse the feature map.The experimental results show that on 2722 image test sets,the accuracy of Fast R-CNN model is 62.7%,and the accuracy of this network model is 81.4%,an increase of 18.7%.Therefore,compared with other network models,the Fast R-CNN model in this paper has stronger feature extraction ability and generalization ability,which improves the technical reference for similar small target detection.

关 键 词：Faster R-CNN 铝型材 缺陷识别 

分 类 号：TP391.4[自动化与计算机技术—计算机应用技术]